presentmanageengine

ARES:AnAnti-jammingREinforcementSystemfor802.
11NetworksKonstantinosPelechrinisDept.
ofCS&EUCRiversidekpele@cs.
ucr.
eduIoannisBroustisDept.
ofCS&EUCRiversidebroustis@cs.
ucr.
eduSrikanthV.
KrishnamurthyCS&E,UCRiversidekrish@cs.
ucr.
eduChristosGkantsidisMicrosoftResearchCambridge,UKchrisgk@microsoft.
comABSTRACTDense,unmanaged802.
11deploymentstemptsaboteursintolaunchingjammingattacksbyinjectingmaliciousinterfer-ence.
Nowadays,jammerscanbeportabledevicesthattransmitintermittentlyatlowpowerinordertoconserveenergy.
Inthispaper,werstconductextensiveexperi-mentsonanindoor802.
11networktoassesstheabilityoftwophysicallayerfunctions,rateadaptationandpowercon-trol,inmitigatingjamming.
Inthepresenceofajammerwendthat:(a)theuseofpopularrateadaptationalgorithmscansignicantlydegradenetworkperformanceand,(b)ap-propriatetuningofthecarriersensingthresholdallowsatransmittertosendpacketsevenwhenbeingjammedandenablesareceivercapturethedesiredsignal.
Basedonourndings,webuildARES,anAnti-jammingREinforcementSystem,whichtunestheparametersofrateadaptationandpowercontroltoimprovetheperformanceinthepresenceofjammers.
ARESensuresthatoperationsunderbenignconditionsareunaected.
TodemonstratetheeectivenessandgeneralityofARES,weevaluateitinthreewirelesstestbeds:(a)an802.
11nWLANwithMIMOnodes,(b)an802.
11a/gmeshnetworkwithmobilejammersand(c)an802.
11aWLANwithTCPtrac.
WeobservethatARESimprovesthenetworkthroughputacrossalltestbedsbyupto150%.
CategoriesandSubjectDescriptorsC.
2.
0[General]:SecurityandProtection;C.
2.
3[ComputerCommunicationNetworks]:NetworkOperationsGeneralTermsDesign,Experimentation,Measurement,Performance,SecurityThisworkwasdonepartiallywithsupportfromtheUSArmyResearchOceundertheMulti-UniversityResearchInitiative(MURI)grantsW911NF-07-1-0318andtheNSFNeTS:WN/Cybertrustgrant0721941.
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforprotorcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationontherstpage.
Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspecicpermissionand/orafee.
CoNEXT'09,December1–4,2009,Rome,Italy.
Copyright2009ACM978-1-60558-636-6/09/12.
.
.
$10.
00.
KeywordsIEEE802.
11,RateControl,PowerControl,Jamming,MIMO1.
INTRODUCTIONThewidespreadproliferationof802.
11wirelessnetworksmakesthemanattractivetargetforsaboteurswithjam-mingdevices[1,2,3,4];thismakesthedefenseagainstsuchattacksverycritical.
Ajammertransmitselectromagneticenergytohinderlegitimatecommunicationsonthewirelessmedium.
Ajammingattackcancausethefollowingeectsinan802.
11network:(a)Duetocarriersensing,co-channeltransmittersdefertheirpackettransmissionsforprolongedperiods.
(b)Thejammingsignalcollideswithlegitimatepacketsatreceivers.
Frequencyhoppingtechniqueshavebeenpreviouslyproposedforavoidingjammers[5][6].
Suchschemeshowever,arenoteectiveinscenarioswithwide-bandjammers[7,8].
Furthermore,giventhat802.
11oper-atesonrelativelyfewfrequencychannels,multiplejammingdevicesoperatingondierentchannelscansignicantlyhurtperformanceinspiteofusingfrequencyhopping[9].
Inthispaper,weaskthequestion:Howcanlegacy802.
11devicesalleviatetheeectsofajammerthatresidesonthesamechannelusedbyalegitimatecommunicatingpair,inrealtimeWeaddressthischallengebydevelopingARES1,anovelmeasurementdrivensystem,whichdetectsthepres-enceofjammersandinvokesrateadaptationandpowercontrolstrategiestoalleviatejammingeects.
Clearly,notmuchcanbedonetomitigatejammerswithunlimitedre-sourcesintermsoftransmissionpowerandspectrume-ciency.
Notehoweverthatinapluralityofcasesthejammingdevicecanberesourceconstrained,withcapabilitiessimilartothatofthelegitimatedevice2.
Portable,battery-operatedjammersaretypicallyconguredtotransmitintermittentlyandsometimesatlowpower,inordertoconserveenergyandharmthenetworkforextendedperiodsoftime.
Inaddition,miscongurationof"legitimate"devicescantransformthemtoresource-constrainedjammers[3].
Insuchcases,AREScaneectivelyghtagainstthemaliciousentity,aswedis-cusslater.
Ourcontributionsarethefollowing:1ARES[pron.
"aris"]wasthegodofwarinGreekmythol-ogy;wechoosethenameasasymbolofthecombatwithjammers.
2Weimplementajammingutilityonacommodity802.
11NICasdescribedinmoredetailinSection3.
1.
Understandingtheimpactofjammersinan802.
11networkwithrate/powercontrol.
First,weperformanin-depthmeasurement-basedexperimentalstudyonourindoortestbed,toquantifytheimpactofjammingwhenemployingrateand/orpowercontrol.
Tothebestofourknowledge,therearenosuchstudiestodate.
Withratecontrol,atransmittercanincreaseorloweritstransmissionratedependingontheobservedpacketdeliveryratio(PDR)atthereceiver.
Withpowercontrol,nodesmayincreasetheirtransmissionpowersand/orclearchannelassessment(CCA)thresholds[10]inordertoincreasetheprobabilityofsuccessfulpacketreception.
ThedesignofARESisdrivenbytwokeyexperimentalobservations:i)Rateadaptationcanbecounter-productive:Inthepresenceofajammerthatisactiveintermittently(andsleepsinbetween),theuseofrateadaptationisnotalwaysbene-cial.
Weconductexperimentswiththreepopularrateadap-tationalgorithms:SampleRate[11],Onoe[12]andAMRR(AdaptiveMultiRateRetry)[13].
Witheveryscheme,weobservethattheuseofrateadaptationmayworkinfavorofthejammer.
Thisisbecause,rateadaptationwastesalargeportionofajammer'ssleepingtimeinordertogradu-allyconvergetothe"best"rate.
Weanalyticallydeterminewhenxedrateoperationsmaybepreferabletotheuseofrateadaptation.
ii)Tuningthecarriersensethresholdisbenecial:WecollectthroughputmeasurementswithmanydierenttransmissionpowersandCCAthresholds.
Wendthat:(a)Inthepresenceofajammer,legitimatetransmissionswithmaximumpowercouldleadtosignicantbenets,onlywhenoperatingatlowdatarates.
(b)IncreasingtheCCAthresholdcanallowatransmitterthatisbeingjammedtosendpacketsandinaddition,facilitatethecaptureofpack-etsinthepresenceofjamminginterference;together,theseeectscansignicantlyreducethethroughputdegradation.
2.
DesigningARES,anovelanti-jammingsystem.
TheaboveobservationsdrivethedesignofARES.
ARESpri-marilyconsistsoftwomodules.
Theratecontrolmodulechoosesbetweenxed-rateassignmentandrateadaptation,basedonchannelconditionsandthejammercharacteristics.
Theprimaryobjectiveofthismoduleistoeectivelyutilizetheperiodswhenajammerisasleep.
ThepowercontrolmoduleadjuststheCCAthresholdtofacilitatethetrans-missionandthereception(capture)oflegitimatepacketsduringjamming.
Careistakentoavoidstarvationofnodesduetothecreationofasymmetriclinks[10].
Thismoduleisusedtofacilitatesuccessfulcommunicationswhilethejam-merisactive.
Althoughrateandpowercontrolhavebeenproposedasinterferencealleviationtechniques,theirbehav-iorhasnotbeenstudiedinjammingenvironments.
Toourknowledge,ourworkisthersttoconductsuchastudy.
3.
ImplementingandexperimentallyvalidatingARES.
WeimplementandevaluatethemodulesofARESonrealhardware,therebymakingARESoneofthefewanti-jammingsystemimplementationsfor802.
11networks.
ARESreliesontheexistenceofanaccuratejammingdetecionmodule.
Itisbeyondthescopeofourworktodesignanewde-tectionscheme,andthusweincorporateamechanismpro-posedpreviouslyin[14].
Todemonstratetheeectivenessandgeneralityofoursystem,weapplyitonthreedier-entexperimentalnetworks:an802.
11nWLANwithMIMOenablednodes,an802.
11a/gmeshnetworkwithmobilejam-mers,andastatic802.
11aWLANwithuplinkTCPtrac.
OurmeasurementsdemonstratethatARESprovidesper-formancebenetsinallthethreenetworks;throughputim-provementsofupto150%areobserved.
2.
BACKGROUNDANDRELATEDWORKInthissection,rstwebrieydescribetheoperationsofajammeranditsattackcapabilities.
Next,wediscussrelevantpreviousstudies.
TypesofJammingAttacks.
Jammerscanbedistin-guishedintermsoftheirattackstrategy;adetaileddiscus-sioncanbefoundin[14].
Non-stopjamming:Constantjammerscontinuouslyemitelectromagneticenergyonachannel.
Nowadays,constantjammersarecommerciallyavailableandeasytoobtain[1,7].
Whileconstantjammersemitnon-decipherablemessages,deceptivejammerstransmitseeminglylegitimateback-to-backdummydatapackets.
Hence,theycanmisleadothernodesandmonitoringsystemsintobelievingthatlegitimatetracisbeingsent.
IntermittentJamming:Asthenamesuggests,thesejammersareactiveintermittently;theprimarygoalistoconservebatterylife.
Arandomjammertypicallyalternatesbetweenuniformly-distributedjammingandsleepingperi-ods;itjamsforTjsecondsandthenitsleepsforTsseconds.
Areactivejammerstartsemittingenergyonlyifitdetectstraconthemedium.
Thismakesthejammerdiculttodetect.
However,implementingreactivejammerscanbeachallenge.
Attackersaremotivatedintousingarandomjammerbe-causeputtingthejammertosleepintermittentlycanin-creaseitslifetimeanddecreasetheprobabilityofdetection[14].
Furthermore,itisthemostgeneralizedrepresentationofajammer;appropriatelychoosingthesleeptimescouldturnthejammerintoaconstantjammeror(withhighprob-ability)areactivejammer.
Moreover,reactivejammersarenoteasilyavailablesincetheyarehardertoimplementandrequirespecialexpertiseonthepartoftheattacker.
Relatedwork.
Mostpreviousstudiesemployfrequencyhoppingtoavoidjammers.
Frequencyhopping,however,cannotalleviatetheinuenceofawide-bandjammer[7,8],whichcaneectivelyjamalltheavailablechannels.
Inad-dition,recentstudieshaveshownthatafewcleverlyco-ordinated,narrow-bandjammerscanpracticallyblocktheentirespectrum[9].
Thus,ARESdoesnotrelyonfrequencyhopping.
Forasetofrelatedstudiesbasedonfrequencyhopping,pleasesee[5],[6],[15].
Xuetal.
[14]developecientmechanismsforjammerdetectionatthePHYlayer(forallthe4typesofjam-mers).
However,theydonotproposeanyjammingmiti-gationmechanisms.
In[16],thesameauthorssuggestthatcompetitionstrategies,wheretransceiversadjusttheirtrans-missionpowersand/oruseerrorcorrectioncodes,mightal-leviatejammingeects.
However,theyneitherproposeananti-jammingprotocolnorperformevaluationstovalidatetheirsuggestions.
LinandNoubir[17]presentananalyti-calevaluationoftheuseofcryptographicinterleaverswithdierentcodingmechanismstoimprovetherobustnessofwirelessLANs.
In[18],theauthorsshowthatintheab-senceoferror-correctioncodes(aswith802.
11)thejammercanconservebatterypowerbydestroyingonlyaportionofalegitimatepacket.
Noubir[19]alsoproposestheuseofacombinationofdirectionalantennaeandnode-mobilityinordertoalleviatejammers.
AREScaneasilybeusedincon-junctionwithdirectionalantennaeorwitherrorcorrectioncodes.
3.
EXPERIMENTALSETUPInthissection,wedescribeourwirelesstestbedandex-perimentalmethodology.
TestbedDescription:Ourtestbedconsistsof37Soekrisnet4826nodes[20],whichmountaDebianLinuxdistribu-tionwithkernelv2.
6,overNFS.
ThirtyofthesenodesareeachequippedwithtwominiPCI802.
11a/gWiFicards,anEMP-86026GwithAtheroschipsetandanIntel-2915.
Theother7nodesareequippedwithoneEMP-86026GandoneRT2860cardthatsupportsMIMO-based(802.
11n)commu-nications.
WeusetheMadWidriver[21]fortheEMP-86026Gcards.
WehavemodiedtheLinuxclientdriver[22]oftheRT2860toenableSTBC(SpaceTimeBlockCoding)support.
Weuseaproprietaryversionoftheipw2200AP(accesspoint)andclientdriver/rmwareoftheIntel-2915card.
WiththisversionweareabletotunetheCCAthresh-oldparameter.
ExperimentalSettingsandMethodology:Weex-perimentwithdierentrateadaptationalgorithmsinthepresenceofrandomjammers.
Wealsoperformexperimentswithvarioustransmissionpowersofjammersandpowers/CCAthresholdsoflegitimatenodes.
Ourmeasurementsencom-passanexhaustivesetofwirelesslinks,routesofdierentlengths,aswellasstaticandmobilejammers.
Weexam-inebothSISOandMIMOlinks.
Weexperimentwiththreemodesofoperation:802.
11a/g/n(unlessotherwisestatedthroughoutthispaper,ourobservationsareconsistentforallthreemodesofoperation).
Theexperimentsareper-formedlateatnightinordertoisolatetheimpactofthejammersbyavoidinginterferencefromco-locatedWLANs.
Bydefault,alldevices(legitimatenodesandjammers)settheirtransmissionpowersto18dBm.
Implementingarandomjammer:Ourimplementa-tionofajammerisbasedonaspecicconguration(CCA=0dBm)andauserspaceutilitythatsendsbroadcastpacketsasfastaspossible.
BysettingtheCCAthresh-oldtosuchahighvalue,weforcethedevicetoignorealllegitimate802.
11signalsevenaftercarriersensing;packetsarriveatthejammer'scircuitrywithpowerslessthan0dBm(evenifthedistancesbetweenthejammerandthelegitimatetransceiversareverysmall).
Weimplementarandomjam-merbutbysettingthesleeptimetozero,itcanfunctionasaconstantjammer.
Weuseasetof4nodesasjammersonourtestbed;theseareequippedwithIntel-2915cardswhichallowCCAtuning.
Traccharacteristics:Weutilizetheiperfmeasure-menttooltogenerateUDPdatatracamonglegitimatenodes;thepacketsizeis1500bytes.
Thedurationofeachexperimentis1hour.
Foreachexperiment,werstenableiperftracbetweenlegitimatenodes,andsubsequently,weactivatethejammer(s).
WeconsiderbothmeshandWLANconnectivity.
Weexperimentwithdierentjammerdistri-butions,namely:(a)frequentjammers,whichareactiveal-mostallofthetime,(b)rarejammers,whichspendmostoftheirtimesleeping,and(c)balancedjammersthathavesim-ilaraveragejammingandsleepingtimes.
WehavedisabledRTS/CTSmessageexchangethroughoutourexperiments(acommondesigndecisioninpractice[23]).
4.
DERIVINGSYSTEMGUIDELINESInthissection,wedescribeourexperimentstowardsun-derstandingthebehavioraltrendsofpowerandrateadap-tationtechniques,inthepresenceofjammer(s).
Ourgoalistodetermineiftherearepropertiesthatcanbeexploitedinordertoalleviatejammingeects.
Weperformexperimentsonbothsingle-hopandmulti-hopcongurations.
4.
1RateAdaptationinJammingEnvironmentsRateadaptationalgorithmsareutilizedtoselectanap-propriatetransmissionrateasperthecurrentchannelcon-ditions.
Asinterferencelevelsincrease,lowerdataratesaredynamicallychosen.
Sincelegitimatenodesconsiderjam-mersasinterferers,rateadaptationwillreducethetrans-missionrateonlegitimatelinkswhilejammersareactive.
Hence,onecouldpotentiallyarguethatratecontrolonle-gitimatelinksincreasesreliabilitybyreducingrateandthus,canprovidethroughputbenetsinjammingenvironments.
Toexaminethevalidityofthisargument,weexperimentwiththreedierent,popularrateadaptationalgorithms,SampleRate[11],AMRR[13]andOnoe[12].
Thesealgo-rithmsarealreadyimplementedontheMadWidriverthatweuse.
Forsimplicity,werstconsiderabalancedrandomjammer,whichselectsthesleepdurationfromauniformdis-tributionU[1,8]andthejammingdurationfromU[1,5](inseconds).
Detailsontheexperimentalprocess:Weperformex-perimentswithbothsingle-hopandmulti-hopcongura-tions.
Ineachexperiment,werstloadtheparticularrate-controlLinux-kernelmodule(SampleRate,AMRRorOnoe)onthewirelesscardsoflegitimatenodes.
Weinitiatedatatracbetweenthenodesandactivatethejammerafterarandomtime.
Wecollectthroughputmeasurementsoneachdatalinkonceevery500msec.
Weusethefollowingtermi-nology:1)FixedtransmissionrateRf:Thisisthenominaltrans-missionrateconguredonthewirelesscard.
2)SaturatedrateRs:ItistherateachievedwhenRfischosentobetherateonthewirelesscard.
Inordertocom-puteRs,foragivenRf,weconsiderlinkswherethepacketdeliveryratio(PDR)is100%fortheparticularsettingofRf;wethenmeasuretherateachievedinpractice.
Weno-ticethatforlowervaluesofRf,thespeciedrateisactuallyachievedonsuchlinks.
However,forhighervaluesofRf(asanexampleRf=54Mbps),theachieveddatarateismuchlower;thishasbeenobservedinotherworke.
g.
[24].
Table1containsamapping,derivedfrommeasurementsonourtestbed,betweenRfandRs.
3)ApplicationdatarateRa:Thisistherateatwhichtheapplicationgeneratesdata.
Rf69121824364854Rs69121824262727Table1:Thesaturated-throughputmatrixinMbps.
Itisdicult(ifnotimpossible)toapriorideterminethebestxedrateonalink.
Giventhis,weset:Rf={minRf:Rf≥Ra},whichisthemaximumratethatisrequiredbytheapplica-tion(wediscusstheimplicationsofthischoicelater).
Ourkeyobservationsaresummarizedbelow:Rateadaptationalgorithmsperformpoorlyonhigh-qualitylinksduetothelongtimesthattheyincurforconvergingtotheappropriatehighrate.
Onlosslesslinks,thexedrateRfisbetter,whilerateadaptationisbenecialonlossylinks.
Wedeferdeningwhatconstitutelosslessorlossylinkstolater;conceptually,weconsiderlosslesslinkstobethoselinksthatcanachievehigherlong-termthroughputsusingaxedtransmissionrateRf,ratherthanbyapplyingrateadaptation.
4.
1.
1Single-hopCongurationsOurexperimentswithone-hopconnectivityinvolve80setsofsender-receiverpairsandonejammerperpair.
Weimposethatajammerinterfereswithonelinkatatimeandthatthelegitimatedatalinksdonotinterferewitheachother.
Thus,weperform20dierentsetsofexperiments,with4isolateddatalinksand4jammersineachexperiment.
Rateadaptationconsumesasignicantpartofthejammer'ssleeptime,toconvergetotheappropriaterate:Assoonasthejammer"goestosleep",thelinkqualityimprovesandthus,theratecontrolalgorithmstartsincreas-ingtherateprogressively.
However,sincethepurposeofajammingattackistocorruptasmanytransmissionsaspos-sible,thejammerwilltypicallynotsleepforalongtime.
Insuchacase,thesleepdurationofthejammerwillnotbeenoughfortheratecontroltoreachthehighestratepossi-ble.
Toillustratethiswechoosetwolinksonourtestbed,onethatcansupport12Mbpsandtheotherthatcansup-port54Mbps.
Figure1depictstheresults.
Weobservethat(a)irrespectiveofwhetherSampleRateoraxedratestrategyisused,duringjammingthethroughputdropstovaluesclosetozerosincethejammerblocksthemediumforthesender,and(b)thethroughputachievedwithSampleR-ateisquitelow,andmuchlowerthanifwextheratetotheconstantvalueof12Mbps.
NotethatwehaveobservedthesamebehaviorwithAMRRandOnoe.
Fixedrateassignmentoutperformsrateadapta-tiononlosslesslinks:Asalludedtoabove,inordertondthebestrateonalinkafteraperiodwherethereisnothroughputduetoajammer,therateadaptationmecha-nismsgraduallyincreasetherate,invokingtransmissionsatallthelowerratesinterim,untilthebestrateisreached.
Forlinksthatcaninherentlysupporthighrates,thispro-cessmightconsumethesleepperiodofthejammer(assug-gestedbytheresultsinFigure1).
Ifthebestrateforalinkwasknownapriori,attheinstancethatthejammergoestosleep,transmissionsmaybeinvokedatthatrate.
Thiswouldutilizethesleepperiodofthejammermoreeectively.
AsobservedinFigure2,thethroughputsachievedwithxedrateassignmentaremuchhigherthanthoseachievedwithrateadaptationonsuchlinks.
Determiningtherighttransmissionratepolicy:ImplicationsofsettingRf={minRf:Rf≥Ra}:Sincetheapplicationdoesnotrequirethelinktosustainahigherrate,thehighestthroughputforthatapplicationrateisreachedeitherwiththischoiceofRforwithsomeratethatislowerthanRa.
Iftherateadaptationalgorithmcon-vergestoaratethatresultsinathroughputthatishigherthanwiththechosenRf,thentheadaptiveratestrategyshouldbeused.
Ifinstead,duringthejammer'ssleeppe-riod,therateadaptationtechniqueisunabletoconvergetosucharate,thexedratestrategyisbetter.
Analyticallydeterminingtherightrate:Inordertodeter-minewhetheritisbettertouseaxedoranadaptive-rateapproachforagivenlink,weperformananalysisbasedonthefollowingparameters:1.
Thedistributionofthejammer'sactiveandsleeppe-riods(wecallthisthejammer'sdistribution).
2.
Theapplicationdatarate,Ra.
3.
Theperformancemetricontheconsideredlegitimatelink,i.
e.
,PDR,linkthroughput,etc.
4.
Therateadaptationschemethatisemployed,i.
e.
,Onoe,SampleRate,etc.
Thekeyscheme-specicfactoristhetransitiontimefromalowerratetothenexthigherrate,underconduciveconditions.
5.
TheeectivenessofthejammerF,measuredbytheachievablethroughputwhilethejammerison.
Thelowerthethroughput,themoreeectivethejammer.
Letussupposethattheexpectedsleepingdurationofthejammerduringacycle,isgivenbyE[ts]andtheexpectedperiodforwhichitisactive,byE[tj].
TheexpecteddurationofacycleisthenE[ts]+E[tj].
Asanexample,ifthejammerpicksitssleepingperiodfromauniformdistributionU[a,b]anditsjammingperiodfromU[c,d],E[ts]andE[tj]areequaltob+a2andd+c2,respectively.
Forsimplicityletusassumethatthelink-qualitymetricemployed3isthePDR.
WithapplicationdatarateRaandxedtransmissionrateRf,thethroughputachievedduringajammer'scycleis:Tfixed=E[ts]E[ts]+E[tj]·PDRf·Rs+E[tj]E[ts]+E[tj]·F,(1)wherePDRfisthePDRofthelinkatrateRf.
RecallthattherateachievedinpracticewithaspeciedrateRfisRs.
Tocomputethethroughputwithrateadaptation,wepro-ceedasfollows.
Letusassumethatx(F,Rs)correspondstotheconvergencetimeoftherateadaptationalgorithm(spe-cictothechosenalgorithm).
Weconsiderthefollowingtwocases.
1)x(F,Rs)Tadaptareclassiedaslosslesslinks.
Clearlyforlossylinksitisbettertousetherateadaptationalgorithm.
Theanal-ysiscanbeusedtocomputePDRTHf,athresholdvalueofPDRfbelowwhich,arateadaptationstrategyperformsbetterthanthexedrateapproach.
Inparticular,bysettingTfixed=Tadaptandsolvingthisequation,onecancomputePDRTHf.
Basedonthis,adecisioncanbemadeonwhethertoenablerateadaptationorusexed-rateassignment.
IftheobservedPDRislargerthanthecomputedthreshold,xedrateshouldbeused;otherwise,rateadaptationshouldbeused.
Validationofouranalysis:Inordertovalidateouranalysis,wemeasurePDRTHfon80dierentlinksinthepresenceofabalancedjammer.
WethencomparethemagainstthePDRTHfvaluescomputedwithouranalysis.
Noteherethattheanalysisitselfdependsonmeasuredval-uesofcertainquantities(suchasthejammerdistributionandthefunctiony(Ri)).
Inthisexperiment,weconsidertheSampleRatealgorithm,andmeasurethevaluesofx(F,Rs)andy(Ri).
Thejammer'ssleeptimefollowsU[0,4]andthejammingtimefollowsU[1,6].
Figure3plotsthevaluesoffunctionyfordierentvaluesofRf.
InTable2,wecomparethetheoreticallycomputedPDRthresholdswiththeonesmeasuredonourtestbed,forvar-iousvaluesofRf.
WeobservethatthePDRfthresholdscomputedwithouranalysisareverysimilartotheonesmea-suredonourtestbed.
Thereareslightdiscrepanciessinceouranalysisisbasedonusingmeasuredaveragevalueswhichmaychangetosomeextentovertime.
Wewishtostressthatwhileweverifyouranalysisassumingthatthejammerisac-tiveandidleforuniformlydistributedperiodsoftime,ouranalysisdependsonlyonexpectedvaluesandisthereforevalidforotherjammerdistributions.
Finally,Figure4showstheadvantageofusingaxedrateapproachoverSampleR-ateforvariousPDRvaluesandwithRf=54Mbps.
WeobservethatSampleRateprovideshigherthroughputsonlyforverylowPDRvalues.
Next,weconsidertwoextremecasesofjamming:frequentandrarejammers(seesection3).
ThedistributionsthatweuseinourexperimentsforthesejammersareshowninTable3.
Notethatbychoosingthejammer'ssleepingandjammingtimefromdistributionslikethatofthefrequentjammer,weessentiallyconstructaconstantjammer.
Withfrequentjam-mers,thedierenceintheperformancebetweenxedrateassignmentandrateadaptationislarger,whileforararejammeritissmaller.
Thisisbecausewithrarejamming,RfMeasuredPDRTHfAnalyticalPDRTHf60.
820.
8390.
520.
55120.
400.
41180.
260.
27240.
190.
21360.
190.
20480.
170.
185540.
150.
185Table2:PDRfthresholdsrateadaptationhasmoretimetoconvergeandthereforeof-tensucceedsinachievingthehighestratepossible;oneob-servestheoppositeeectwhenwehaveafrequentjammer.
TheresultsareplottedinFigures5and6.
-Sleeptime(sec)Jammingtime(sec)BalancedU[1,8]U[1,5]RareU[1,5]U[1,2]FrequentU[1,2]U[1,15]Table3:Thejammingdistributionsthatweuseinourexperiments.
4.
1.
2RandomJamminginMulti-hopTopologiesNext,weexaminetheimpactofarandomjammerontheend-to-endthroughputofamulti-hoppath.
Weexperimentwith15dierentroutesonourtestbed.
Wexstaticroutesofvariouslengths(from2to4linksperroute)utilizingtherouteUnixtoolinordertomodifytheroutingtablesofnodes.
Weplaceajammersuchthatitaectsoneormorelinks.
Alongeachroute,linksthatarenotaectedbythejammerconsistentlyusearateadaptationalgorithm.
Onthelinksthataresubjecttojamming,ouranalysisdictatesthedecisiononwhethertousexedoradaptiverateassign-ment.
Wemeasuretheend-to-endthroughputontheroute.
Weshowourresultsforroutesonwhich,intheabsenceofajammer,end-to-endthroughputsof6and12Mbpswereobserved.
FromFigure7weseethatthebehaviorwithrateadaptationonmulti-hoproutes,inthepresenceofarandomjammer,isthesameasthatonasingle-hoplink.
Inpartic-ular,withlowdatarates,asucientlyhighPDRhastobesustainedovertheroute,inorderforaxedrateapproachtoperformbetterthanrateadaptation.
Ontheotherhand,whenroutessupporthighdatarates,xingtherateontheindividuallinks(thatareaectedbythejammer)asperour05101520253002468101214RateTime(sec)6Mbps9Mbps12Mbps18Mbps24Mbps36Mbps48Mbps54MbpsFigure3:Measuredconver-gencetimesoftheMadWiSampleRatealgorithm,forthedierentapplicationdatarates.
-505101500.
20.
40.
60.
81Throughputgain(Mbps)PDRFigure4:ThroughputgainofxedrateVs.
SampleRate,forvariouslinkqualitiesandforapplicationdatarateof54Mbps.
05101520253054483624181296AverageThroughput(Mbps)Rate(Mbps)FixedRateSampleRateFigure5:Theperformancewithrarejammersisalignedwithourobservationsforthecasewithbalancedjammers.
(Ra=Rf)01234567854483624181296AverageThroughput(Mbps)Rate(Mbps)FixedRateSampleRateFigure6:Fixedrateimprovestheperformancemorethanrateadaptationathighrates,withfrequentjammers.
(Ra=Rf)012345126AverageThroughput(Mbps)Rate(Mbps)FixedRateSampleRateFigure7:Rateadaptationpresentsthesamebehaviorinmultihoplinks;itprovideslowerthroughputathighrates.
00.
20.
40.
60.
81(18,5)(18,18)Throughput(%)sustained(PLdBm,PJdBm)6Mbps54MbpsFigure8:Percentageoftheisolatedthroughput,forvari-ousPLandPJcombinations,fortwodierenttransmissionrates.
analyticalframework,provideshigherbenets.
Choosingtherightpolicyinpractice:Tosumma-rizeourndings,ouranalysisdemonstratesthatusingaxedratemaybeattractiveonlosslesslinkswhileitwouldbebettertouserateadaptationonlossylinks.
However,asdiscussed,determiningwhentouseoneovertheotherinrealtimeduringsystemoperationsisdicult;thede-terminationrequirestheknowledgeofx(F,Rs),y(Ri)andestimatesofhowoftenthejammerisactive/asleep,onav-erage.
Thus,wechooseasimplerpracticalapproachthatwecallMRCforMarkovianRateControl.
WewilldescribeMRCindetaillater(insection5)butinanutshell,MRCinducesmemoryintothesystemandkeepstrackofthefeasi-bleratesduringbenignjamming-freeperiods;assoonasthejammergoestosleep,legitimatetransmissionsareinvokedatthemostrecentrateusedduringtheprevioussleepingcy-cleofthejammer.
Wealsoperformoinemeasurementsbydirectlyusingouranalyticalformulation(withknowledgeoftheaforementionedparameters);thesemeasurementsserveasbenchmarksforevaluatingtheecacyofMRC(discussedinsection6).
4.
2PerformanceofPowerControlinthePres-enceofRandomJammingNext,weexaminewhethertuningpowerlevelscanhelpcopewiththeinterferenceinjectedbyajammer.
Ifwecon-siderasinglelegitimatedatalinkandajammer,increment-ingthetransmissionpoweronthedatalinkshouldincreasetheSINR(signal-to-interferenceplusnoiseratio)ofthere-ceiveddatapackets.
Thus,onecouldarguethatincreasingthetransmissionpowerisalwaysbenecialinjammingen-vironments[17].
Wevarythetransmissionpowersofboththejammerandlegitimatetransceiver,aswellastheCCAthresholdofthelatter.
Notethatthejammer'stransmissiondistributionisnotveryrelevantinthispartofourstudy.
Ourexpectationisthattuningthepoweroflegitimatetransceiverswillpro-videbenetswhilethejammerisactive.
Inotherwords,onecanexpectthatthebenetsfrompowercontrolwillbesimilarwithanytypeofjammer.
Wedenethefollowing:RSSITR:TheRSSIofthesignalofthelegitimatetransmitteratitsreceiver.
RSSIRT:TheRSSIofthesignalinthereversedirec-tion(thereceiverisnowthetransmitter).
RSSIJTandRSSIJR:TheRSSIvaluesofthejam-mingsignalatthelegitimatetransmitterandreceiver,respectively.
RSSIJ:Theminimumof{RSSIJT,RSSIJR}.
PLandCCAL:ThetransmissionpowerandtheCCAthresholdatlegitimatetransceivers.
PJ:Thetransmissionpowerofthejammer.
Ourmainobservationsarethefollowing:MitigatingjammingeectsbyincrementingPLisviableatlowdatarates.
Itisextremelydif-culttoovercomethejamminginterferenceathighrates,simplywithpoweradaptation.
IncreasingCCALrestores(inmostcases)theisolatedthroughput(thethroughputachievedintheabsenceofjammers).
Wepresentourexperimentsandtheinterpretationsthereof,inwhatfollows.
4.
2.
1IncreasingPLtocopewithjamminginterfer-enceIncreasingPLwillincreasetheSINRandonemightexpectthatthiswouldreducetheimpactofjamminginterferenceonthethroughput.
Inourexperimentswequantifythegainsfromemployingsucha"brute-force"approach.
Detailsontheexperimentalprocess:Weperformmea-surementson80dierentlinksandwith4jammers.
Wecon-siderdierentxedvaluesforPJ(from1dBmto18dBm).
ForeachofthesevalueswevaryPLbetween1and18dBmandobservethethroughputinthepresenceofthejammer,forallpossiblexedtransmissionrates.
Foreachchosenpairofvalues{PL,PJ},werun60-minuterepeatedexperimentsandcollectanewthroughputmeasurementonceevery0.
5seconds.
Bothend-nodesofalegitimatelinkusethesametransmissionpower.
ThecombinationofhighPLandlowdataratehelpsmitigatetheimpactoflow-powerjammers.
Weexperimentwithmanydierentlocationsofthejam-mers.
Ourmeasurementsindicatethatwhenhightransmis-sionratesareused,increasingPLdoesnothelpalleviatetheimpactofjammers.
SampleresultsaredepictedinFig-ure8.
Inthisgure,weplotthepercentageoftheisolatedthroughputachievedinthepresenceofjamming,fortworep-resentativecombinationsofPLandPJandfor2dierentrates.
Inourexperimentsonthe80consideredlinks,therewerenolinkswhereincrementingPLincreasedthethrough-putathighdatarates,evenwithverylowjammingpowers.
WhiletherecouldexistcaseswhereincrementingPLcouldyieldbenetsathighrates,thiswasnotobserved.
Incon-trast,weobservethatwithlowdataratesandwhenPJislow,datalinkscanovercomejammingtoalargeextentbyincreasingPL.
Figure9depictsanotherrepresentativesub-setofourmeasurementresultswherealllegitimatenodesusePL=18dBm,whilePJisvariedbetween1and18dBm.
WeobservethatthecombinationofhighPLwithlowdataratehelpsovercometheimpactofjamming,whenPJislow.
NotealsothatwhenPJishigh,itisextremelydiculttoachievehighaveragethroughput.
Theaboveobservationscanbeexplainedbytakingacare-fullookatthefollowingtwocases:Strongjammer:LetusconsiderajammersuchthatRSSIJ>CCAL.
Thiscanresultintwoeects:(a)Thesenderwillsensethatthemediumisconstantlybusyandwilldeferitspackettransmissionsforprolongedperiodsoftime.
(b)ThesignalsofboththesenderandthejammerwillarriveatthereceiverwithRSSIvalueshigherthanCCAL.
Thiswillresultinapacketcollisionatthereceiver.
Inbothcases,thethroughputisdegraded.
OurmeasurementsshowthatitisnotpossibletomitigatestrongjammerssimplybyincreasingPL.
Weakjammer:Letussupposethatthejammer'ssignalsarrivewithlowRSSIatlegitimatenodes.
Thismaybeei-therduetoenergy-conservationstrategiesimplementedbythejammercausingittouselowPJ(e.
g.
,2dBm),orduetopoorchannelconditionsbetweenajammerandalegitimatetransceiver.
Athightransmissionrates,theSINRrequiredforthesuccessfuldecodingofapacketislargerthanwhatisrequiredatlowrates(showninTable4)[10].
Ourthrough-putmeasurementsshowthateveninthepresenceofweakjammers,theSINRrequirementsathightransmissionratesaretypicallynotsatised.
However,sincetheSINRrequire-mentsatlowerdataratesarelessstringent,thecombinationofhighPLandlowrate,providessignicantthroughputben-ets.
DataRate69121824364854SINR(dB)67.
8910.
81718.
82424.
6Table4:SINRlevelsrequiredforsuccessfulpacketdecoding,in802.
11a/g.
4.
2.
2TuningCCALonsingle-hopsettingsNext,weinvestigatethepotentialofadjustingCCALinconjunctionwithPL.
Implementationandexperimentaldetails:FortheseexperimentsweexclusivelyusetheIntel-2915cards;thesecardsallowustotunetheCCAthreshold.
Wehavemod-iedaprototypeversionoftheAP/clientdriver,inordertoperiodicallycollectmeasurementsforRSSITR,RSSIRTandRSSIJ.
Weconsider80AP-clientdatalinks,withtraf-cowingfromtheAPtotheclient.
Asbefore,wedividethe80datalinksinto20setsof4isolatedlinks.
WeuseIn-tel'sproprietaryrateadaptationalgorithm,whichhasbeenimplementedinthermwareoftheIntel-2915cards.
WemeasuretheachieveddatathroughputfordierentvaluesofPLandCCAL.
BothnodesofadatalinkusethesamepowerandCCAthresholdvalues.
TuningtheCCAthresholdisapotentialjammingmitigationtechnique.
Tobeginwith,weperformthrough-putmeasurementswiththedefaultCCALvalue(-80dBm),andwithvariousRSSIJvalues.
WeobservefromFigure10thatwhenRSSIJCCAL,whileothersignalsareconsideredtobebackgroundnoise.
Moreover,evenwhenRSSIJisslightlylargerthanCCAL,westillobservedecentthroughputachievementsforthecaseswhereindatalinksoperateathighSINRregimes.
ThisisbecausethereportedRSSIvalueisanaverageandthejam-mersignalcouldbebelowthethresholdevenhere,inmanycases.
ThesemeasurementsimplythattheabilitytotuneCCALcanhelpreceivedatapacketscorrectly,evenwhilejammersareactive.
Inordertofurtherexplorethepotentialofsuchanap-proach,wevaryCCALfrom-75to-30dBmoneachoftheconsidered80links.
Figure11depictstheresultsforthecasewhereCCALisequalto-50dBm.
WeobservethatincreasingCCALresultsinsignicantlyhigherdatathroughputs,evenwithquitehighRSSIJvalues.
Morespecically,fromFigure11weobservethatwhenRSSIJislowerthanCCAL,linkscanachieveupto95%ofthethroughputthatisachievedwhenthemediumisjamming00.
20.
40.
60.
81024681012141618Throughput(%)PJ(dBm)6Mbps9Mbps12Mbps18Mbps24Mbps36Mbps48Mbps54MbpsFigure9:Percentageoftheiso-latedthroughputinthepres-enceofabalancedjammerforvariousPJandPJvaluesanddatarates.
Figure10:PercentageoftheisolatedthroughputinthepresenceofabalancedjammerVs.
RSSIJ,forCCAL=–80dBm.
Figure11:Percentageoftheisolatedthroughput,forvari-ousRSSIJvalues,andforCCAL=–50dBm.
00.
20.
40.
60.
81-80-75-70-65-60-55-50-45-40Throughput(%)CCA(dBm)PL20dBmPL15dBmPL10dBmPL05dBmFigure12:Percentageoftheisolatedthroughput,forvari-ousCCALvaluesandvariousPLvalues.
PJ=20dBm.
00.
20.
40.
60.
81-40-80Throughput(%)CCA(dBm)Figure13:CarefulCCAadap-tationsignicantlyimprovestheend-to-endthroughputalongaroute.
051015202554483624181296AverageThroughput(Mbps)Rate(Mbps)FixedRateSampleRateMRCwithK=3MRCwithK=30Figure14:MRCoutperformscurrentrateadaptationalgo-rithms,especiallyforhighval-uesofK.
free.
WhenRSSIJ≈CCAL,datalinksstillachieveupto70%ofthejamming-freethroughput(captureofdatapack-etsisstillpossibletoasignicantextent).
Asonemightexpect,ifRSSIJCCAL,therearenoperformanceben-ets.
Ourobservationsalsoholdinsomescenarioswhere,PJ>PL.
Figure12presentstheresultsfromonesuchscenario.
WeobservethatappropriateCCAsettingscanallowle-gitimatenodestoexchangetraceectively,evenwhenPJPL.
ThisispossibleifthelinkconditionsbetweenthejammerandthelegitimatetransceiversarepoorandresultinlowRSSIJ.
NoteherethatonecannotincreaseCCALtoarbitrarilyhighvaluesonlegitimatenodes.
DoingsoislikelytocompromiseconnectivitybetweennodesordegradethethroughputduetofailureofcapturingpacketsasseeninFigure12forPL=5dBmandPL=10dBm.
4.
2.
3TuningCCALinmulti-hopcongurationsWeperformexperimentswithvariousCCAthresholdsalongaroute.
Previousstudieshaveshownthatinordertoavoidstarvationduetoasymmetriclinks,thetransmis-sionpowerandtheCCAthresholdneedtobejointlytunedforallnodesofthesameconnected(sub)network[10].
Inparticular,theproductC=PL·CCALmustbethesameforallnodes.
Giventhis,weensurethatCisthesameforallnodesthatarepartofaroute.
Inparticular,wesetPLtobeequaltothemaximumpossiblevalueof20dBmonallnodesofaroute;foreachrun,CCAListhereforesettobethesameonallofthenodesontheroute.
Throughoutourexperimentswithmulti-hoptrac,nodesononeroutedonotinterferewithnodesthatareonotherroutes.
Inscenar-ioswherenodesbelongingtodierentroutesinterferewitheachother,ifallnodesusethesamePL,theirCCALvaluesmustbethesame[10],[25].
However,wedidnotexperimentwithsuchscenariosgiventhatourobjectiveistoisolatetheimpactofajammerandnottoexamineinterferencebetweencoexistingsessionsinanetwork.
Weexperimentwiththesamemulti-hopsettingsasinsection4.
1.
2.
Figure13presentstheresultsobservedononeofourroutes.
WeobservethatcarefulCCAtuningcanprovidesignicantaverageend-to-endthroughputbenetsalongaroute.
5.
DESIGNINGARESInthissection,wedesignoursystemARESbasedontheobservationsfromtheprevioussection.
ARESiscomposedoftwomainmodules:(a)aratemodulethatchoosesbe-tweenxedoradaptive-rateassignment,and(b)apowercontrolmodulethatfacilitatesappropriateCCAtuningonlegitimatenodes.
RateModuleinARES:Asdiscussedinsection4.
1,ourexperimentswiththreepopularrateadaptationalgorithmsshowthattheconvergencetimeofthealgorithmsaectsthelinkperformanceinrandom-jammingenvironments.
Thisconvergencetimeislargelyimplementationspecic.
Asanexample,ourexperimentswithbothSampleRateandOnoeshowthatinmanycasesittakesmorethan10secforbothalgorithmstoconvergetothe"best"rate;[26]reportssimilarobservations.
TheratemoduleinARESdecidesonwhetheraxedoranadaptive-rateapproachshouldbeapplied.
MRC:MarkovianRateControl:MRCisanalgorithm–patchthatcanbeimplementedontopofanyratecontrolalgorithm.
MRCismotivatedbyouranalysisinsection4.
However,asdiscussedearlier,itdoesnotdirectlyapplytheanalysis,sincethiswouldrequireextensiveoinemeasure-ments(thecollectionofwhichcanbetime-consuming)andestimatesofthejammeractiveandsleepperiods.
ThekeyideathatdrivesMRCisthatarateadaptationalgorithmneednotexaminetheperformanceatallthetransmissionratesduringthesleepingperiodofthejammer.
Thealgo-rithmsimplyneedstorememberthepreviouslyusedtrans-missionrate,anduseitassoonasthejammergoestosleep.
Simplyput,MRCintroducesmemoryintothesystem.
Thesystemkeepstrackofpasttransmissionratesandhopstothestoredhighest-ratestateassoonasthejammergoestosleep.
Sincethechannelconditionsmayalsochangeduetothevariabilityintheenvironment,MRCinvokesthere-scanningofallratesperiodically,onceeveryKconsecutivesleeping/jammingcycles.
WhenK=1wedonotexpecttohaveanybenets,sincethescanningtakesplaceineachcycle.
NoteherethattheappropriatevalueofKdependsontheenvironmentandthesleepandactiveperiodsofthejammer.
OnecouldadaptivelytunetheKvalue.
Asanexample,anadditiveincreaseadditivedecreasestrategymaybeusedwhereonewouldincreasethevalueofKuntiladegradationisseen.
TheKvaluewouldthenbedecreased.
Theim-plementationofsuchastrategyisbeyondthescopeofthispaperandwillbeconsideredinthefuture.
ImplementationdetailsofMRC:Theimplementation(a)keepstrackofthehighesttransmissionrateusedoverabe-nigntimeperiod(whenthejammerisasleep)and,(b)ap-pliesthisrateimmediatelyuponthedetectionofthenexttransitionfromthejammer'sactiveperiodtothesleepingperiod.
Figure14presentsasetofmeasurementswithMRC,withintermittentSampleRateinvocations(onceeveryKcycles)forK={3,30}.
WeobservethatMRCoutperformspureSampleRateinjammingenvironments,especiallywithlargervaluesofK.
WithsmallK,therateadaptationalgorithmisinvokedoftenandthisreducestheachievedbenets.
Fur-thermore,MRCprovidesthroughputthatisclosetothemaximumachievableonthelink(whichmaybeeitherwithxedoradaptiverate,dependingonwhetherthelinkislossyorlossless).
PowerControlModuleinARES:Asdiscussedinsec-tion4.
2,increasingPLisbenecialatlowrates;whileathighratesthisisnotparticularlyuseful,itdoesnothurteither.
Sinceourgoalinthispaperistoproposemethodsforovercomingtheeectsofjamming(andnotlegitimate)interference,weimposetheuseofthemaximumPLbyallnodesinthepresenceofjammers.
Thedesignofapowercontrolmechanismthatinadditiontakesintoaccounttheimposedlegitimateinterference(duetohighPL)isbeyondthescopeofthispaper.
Moresignicantly,ourpowercontrolmoduleovercomesjamminginterferencebyadaptivelytuningCCAL.
Themodulerequiresthefollowinginputsoneachlink:ThevaluesofRSSITR,RSSIRT,RSSIJR,andRSSIJT.
Thesevaluescanbeeasilyobservedinrealtime.
Anestimationfortheshadowfadingvariationofthechannel,.
Duetoshadowfading,theaboveRSSIvaluescanoccasionallyvaryby.
Thevalueofisdependentontheenvironmentofdeployment.
OnecanperformoinemeasurementsandcongurethevalueofinARES.
WedeterminethevariationsinRSSImeasurementsviaex-perimentsonalargesetoflinks.
Themeasurementsindicatethatisapproximately5dBforourtestbed(alesscon-servativevaluethanwhatisreportedin[27]).
ThevalueofCCALhastobeatleastdBlowerthanbothRSSITRandRSSIRT,toguaranteeconnectivityatalltimes.
Hence,ARESsets:CCAL=min(RSSITR,RSSIRT),ifmax(RSSIJT,RSSIJR)≤min(RSSITR,RSSIRT).
Otherwise,CCALisnotchanged4.
Thisensuresthatlegiti-matenodesarealwaysconnected,whilethejammer'ssignalisignoredtotheextentpossible.
Ourexperimentsindicatethat,especiallyifmax(RSSIJT,RSSIJR)≤min(RSSITR,RSSIRT)2,thedatalinkcanoperateasifitisjamming-free.
Inordertoavoidstarvationeects,thetuningoftheCCAthresholdshouldbeperformedonlywhennodesthatpar-ticipateinpowercontrolbelongtothesamenetwork[25].
UnlesscollocatednetworkscooperateinjointlytuningtheirCCA(asperourscheme),ourpowercontrolmodulewillnotbeused.
Notethatwhenjammingattacksbecomemoreprevalent,cooperationbetweencoexistingnetworksmaybeessentialinordertoghttheattackers.
Hence,insuchcasescollocatednetworkscanhaveanagreementtojointlyin-creasetheCCAthresholdswhenthereisajammer.
Implementationdetails:Ourpowercontrolalgorithmcanbeappliedinacentralizedmannerbyhavingallle-gitimatenodesreporttherequiredRSSIvaluestoacentralserver.
ThecentralserverthenappliesthesameCCALvaluetoallnodes(ofthesameconnectednetwork).
ThechosenCCAListhehighestpossibleCCAthresholdthatguaranteesconnectivitybetweenlegitimatenodes.
Thisre-portingrequirestrivialmodicationsonthewirelessdrivers.
Wehaveimplementedacentralizedfunctionalitywhenournetworkisconguredasamulti-hopwirelessmesh.
Inadistributedsetting,ouralgorithmisapplicableaslongaslegitimatenodesareabletoexchangeRSSIinfor-mation.
EachnodecanthenindependentlydeterminetheCCALvalue.
Todemonstrateitsviability,weimplementandtestadistributedversionofthepowercontrolmoduleina802.
11a/gWLANconguration.
Inparticular,wemod-ifytheIntelprototypeAPdriver,byaddinganextraeldinthe"Beacon"template.
ThisneweldcontainsamatrixofRSSIvaluesofneighboringjammersandlegitimatenodes.
WeenablethedecodingofreceivedbeaconsintheAPdriver(theydonotreadthesebydefault).
Assumingthatajam-merimposesalmostthesameamountofinterferenceonalldevices(APandclients)withinacell,theAPofthecelldeterminesthenalCCALafteraseriesofiterationsinamannerverysimilartotheapproachesin[25],[10].
CombiningthemodulestoformARES:WecombineourrateandpowercontrolmodulestoconstructARESasshowninFigure15.
ThegoalofARESistoapplythein-dividualmodulesasappropriate,oncethejammersarede-tected.
Forthelatter,ARESreliesonalreadyexistingjam-mingdetectionschemesandinheritstheiraccuracy.
Forex-ample,themechanismthatwasproposedin[14]canbeused;thisfunctionalityperformsaconsistencycheckbetweenthe4WechoosenottotuneCCAL,unlesswearecertainthatitcanhelpalleviatejamminginterference.
Figure15:ARES:ourAnti-jammingReinforcementSystem.
instantaneousPDRandRSSIvalues.
IfthePDRisex-tremelylowwhiletheRSSIismuchhigherthanthedefaultCCAL,thenodeisconsideredtobejammed.
Wewanttoreiterate,thatitisbeyondthescopeofourworktodesignanew,evenmoreaccurate,detectionscheme.
ARESappliesthepowercontrolmodulerst,sincewiththismodule,theimpactofthejammer(s)couldbecom-pletelyovercome.
Ifthereceiverisabletocaptureandde-codeallpacketsinspiteofthejammer'stransmissions,nofurtheractionsarerequired.
NotethatevenifCCAL>RSSIJ,thejammercanstillaectthelinkperformance.
ThisisbecausewithCCAtuningthejammingsignal'spowerisaddedtothenoisepower.
Hence,eventhoughthethrough-putmayincrease,thelinkmaynotachievethe"jamming-freeperformance"whilethejammerisactive.
IfthejammerstillhasaneectonthenetworkperformanceaftertuningCCAL,(orifCCAtuningisinfeasibleduetothepresenceofcollocateduncooperativenetworks)ARESenablestheratemodule.
Notethatthetwomodulescanoperateindepen-dentlyandthesystemcanbypassanyofthemincasethehardware/softwaredoesnotsupportthespecicfunctional-ity.
6.
EVALUATINGOURSYSTEMWerstevaluateARESbyexaminingitsperformanceinthreedierentnetworks:aMIMO-basedWLAN,an802.
11meshnetworkinthepresenceofmobile-jammers,andan802.
11aWLANsettingwhereuplinkTCPtracisconsid-ered.
ARESbooststhethroughputofourMIMOWLANunderjammingbyasmuchas100%:Ourobjectivehereistwofold.
First,weseektoobserveandunderstandthebehaviorofMIMOnetworksinthepresenceofjamming.
Second,wewishtomeasuretheeectivenessofARESinsuchsettings.
Towardsthis,wedeployasetof7nodesequippedwithRalinkRT2860miniPCIcards.
Experimentalset-up:WeexaminethecaseforaWLANsetting,sincetheRT2860driverdoesnotcurrentlysupportthead-hocmodeofoperations.
MIMOlinkswithSpace-TimeBlockCodes(STBC)areexpectedtoproviderobust-nesstosignalvariations,therebyreducingtheaverageSINRthatisrequiredforachievingadesiredbiterrorrate,ascom-paredtoacorrespondingSISO(Single-InputSingle-output)link.
Forourexperiments,weconsider2APs,with2and3clientseach,andtwojammers.
Fully-saturateddownlinkUDPtracowsfromeachAPtoitsclients.
ApplyingARESonaMIMO-basedWLAN:WerstrunexperimentswithoutenablingARES.
Interestingly,weobservethatinspiteofthefactthatSTBCisused,802.
11nlinkspresentthesamevulnerabilitiesas802.
11aorglinks.
Inotherwords,MIMOdoesnotoersignicantbenetsbyitself,inthepresenceofajammer.
Thisisduetothefactthat802.
11nisstillemployingCSMA/CAandasaresultthejammingsignalscanrenderthemediumbusyforaMIMOnodeaswell.
Moreover,forSTBCcodestoworkeectivelyandprovideareductionintheSINRforadesiredbiterrorrate(BER),thesignalsreceivedonthetwoantennaelementswillhavetoexperienceindependentmultipathfadingeects.
Inotherwords,alineofsightordominantpathmustbeabsent.
However,inourindoortestbed,giventheproximityofthecommunicatingtransceiverpair,thismaynotbethecase.
Thus,littlediversityisachieved[28]anddoesnotsuceincopingwiththejammingeects.
Next,weapplyARESandobservethebehavior.
ThelogicalsetofstepsthatARESfollows(inFigure15)is1→5→7→8→9.
SincetheCCAthresholdisnottun-ablewiththeRT2860cards,ARESderivesdecisionswithre-gardstoratecontrolonly.
Figure16depictstheresults.
WeobservethatthecongurationwithARESoutperformstherateadaptationschemethatisimplementedontheRT2860cardsinthepresenceofthejammer,byasmuchas100%.
Notethathighergainswouldbepossible,ifARESwasabletoinvokethepowercontrolmodule.
InFigure16wealsocomparethethroughputwithMRCagainstthesuggestedsettingswithouranalysis(theseset-tingsallowustoobtainbenchmarkmeasurementspossiblewithglobalinformation).
Theparametersinputtotheanal-ysisarethefollowing:(a)ThejammerisbalancedwithajammingdistributionU[1,5]andasleepdistributionU[1,6].
(b)WeexaminefourRavalues:13.
5,27,40.
5,54Mbps.
(c)F=0Mbps.
(d)Weinputestimatesofthey(Ri)valueswhichareobtainedviacomprehensiveoinemeasurements.
(e)TheoinemeasuredPDRf.
Weobservethattheper-formancewithMRCisquiteclosetoourbenchmarkmea-surements.
Theseresultsshowthatinspiteofhavingnoinformationwithregardstothejammerdistributionortheconvergencetimesoftherateadaptationalgorithms,MRCisabletosignicantlyhelpinthepresenceofarandomjam-mer.
ARESincreasesthelinkthroughputbyupto150%inan802.
11ameshdeploymentwithmobilejam-mers:Next,weapplyARESinan802.
11ameshnetworkwithmobilejammersandUDPtrac.
Weconsiderafre-quentjammer(jammingdistributionU[1,20]andsleepingdistributionU[0,1]).
Thejammermovestowardsthevicin-ityofthelegitimatenodes,remainsthereforkseconds,andsubsequentlymovesaway.
Forthemobilejammerweusedalaptop,equippedwithoneofourIntelcards,andcarried0510152025303513.
52740.
554AverageThroughput(Mbps)Rate(Mbps)BenchmarkResultsPerformancewithARESPerformancewithoutARESFigure16:ARESprovidessignicantthroughputbenetsinaMIMOnetworkinthepresenceofjammers.
0510152025300100200300400500600700Throughput(Mbps)TimeWithARESWithoutARESFigure17:ARESprovidessignicantthroughputimprove-mentinmobile-jammingscenarios.
051015202530020406080100120Throughput(Mbps)TimeWithARESWithoutARESFigure18:ARESim-provestheclient-APlinkthroughputby130%withTCPtracscenarios.
051015203APs2APs1APAverageAPthroughput(Mbps)NumberofneighborAPsWithoutARESWithARESFigure19:MRCim-provesthethroughputofneighborlegitimatedevices,ascomparedtoSampleRate.
itaround.
Thepowercontrolmoduleisimplementedinacentralizedmanner.
ARESincreasesCCALinordertoover-cometheeectsofjamminginterference,totheextentpossi-ble.
Inthiscase,duetotheaggressivenessoftheconsideredjammer(prolongedjammingduration),therateadaptationmoduledoesnotprovideanybenets(sinceratecontrolhelpsonlywhenthejammerissleeping).
Inthisscenario,ARESfollowsthesteps:1→2→3→4→6→7→8→9.
Figure17depictsthroughput-timetraces,withandwithoutARES,foranarbitrarilychosenlinkandk≈200.
TheuseofAREStremendouslyincreasesthelinkthroughputduringthejammingperiod(byasmuchas150%).
Wehaveob-servedthesamebehaviorwithadistributedimplementationofthepowercontrolmoduleinan802.
11aWLANsetting.
ARESimprovesthetotalAPthroughputbyupto130%withTCPtrac:Next,weapplyARESona802.
11aWLAN.
Forthisexperiment,weusenodesequippedwiththeIntel-2915cards.
Weconsiderasettingwith1APand2clients,whereclientscansenseeachothers'transmis-sions.
Weplaceabalancedjammer(jammingdistributionU[1,5]andsleepingU[1,8])suchthatall3legitimatenodescansenseitspresence.
Weenablefully-saturateduplinkTCPtracfromallclientstotheAP(usingiperf)andwemeasurethetotalthroughputattheAP,onceevery0.
5sec.
Inthisscenario,ARESfollowsthelogicalsteps:1→2→3→4→6→7→8→9.
FromFigure18,weobservethatthetotalAPthroughputisimprovedbyupto130%duringtheperiodsthatthejammerisactive.
Theben-etsarelessapparentwhenthejammerissleepingbecauseTCP'sowncongestioncontrolalgorithmisunabletofullyexploittheadvantagesoeredbythexedratestrategy.
ApplyingMRConanAPimprovesthethrough-putofneighborAPsbyasmuchas23%:WithMRC,ajammednodeutilizesthelowestrate(whenthejammerisactive)andhighestrate(whenthejammerissleeping)thatprovidethemaximumlong-termthroughput.
Withthis,thejammednodeavoidsexaminingtheintermediateratesand,asweshowedabove,thisincreasesthelinkthrough-put.
Wenowexaminehowthisrateadaptationstrategyaectstheperformanceofneighborlegitimatenodes.
Weperformexperimentsonatopologyconsistingof4APsand8clients,with2clientsassociatedwitheachAP,allsetto802.
11amode.
Abalancedjammerwithajammingdistri-butionU[1,5]andasleepdistributionU[1,6]isplacedsuchthataectsonlyoneoftheAPs.
OnlytheaectedAPisrunningMRC;therestoftheAPsuseSampleRate.
WeactivatedierentnumbersofAPsatatime,andween-ablefully-saturateddownlinktracfromtheAPstotheirclients.
Figure19depictstheaveragetotalAPthroughput.
Interestingly,weobservethattheuseofMRConjammedlinksimprovestheperformanceofneighborAPsthatarenotevenaectedbythejammer.
ThisisbecausethejammedAPdoesnotsendanypacketsusingintermediatebitrates(suchaswiththedefaultoperationofrateadaptational-gorithms).
SinceMRCavoidsthetransmissionofpacketsatlower(thatthehighestsustained)bitrates,thejammedAPdoesnotoccupythemediumforasprolongedperiodsaswiththedefaultratecontroltechniques;thetransmissionofpacketsatthehighrate(whilethejammerisasleep)takeslesstime.
Hence,thisprovidesmoreopportunitiesforneigh-borAPstoaccessthemedium,therebyincreasingtheAPthroughput.
Specically,weobservethatthethroughputofoneneighborAPisimprovedby23%(whenthetopologyconsistsofonly2APs,oneofwhichisjammed).
Aswefur-therincreasethenumberofneighborAPs,thebenetsduetoMRCarelesspronounced,duetoincreasedcontention(Figure19).
ARESconvergesrelativelyquickly:Finally,weper-formexperimentstoassesshowquicklythedistributedformofARESconvergestoarateandpowercontrolsetting.
Inanutshell,ourimplementationhasdemonstratedthatthenetwork-wideconvergencetimeofARESisrelativelysmall.
WithMRC,theratecontrolmodulecanveryrapidlymakeadecisionwithregardstotheratesetting;assoonasthejam-merisdetected,MRCappliestheappropriatestoredlowestandhighestrates.
Withregardstotheconvergenceofthepowercontrolmodule,recallthatourimplementationinvolvesthedissem-inationofthecomputedCCAvaluethroughtheperiodictransmissionofbeaconframes(onebeaconframeper100msecistransmittedwithouripw2200driver)[25].
Asonemightexpect,thejammer'ssignalmaycollidewithbeaconframes,andthismakesitmoredicultforthepowercon-trolmoduletoconverge.
Notealsothatasreportedin[25,29],beacontransmissionsarenotalwaystimely,especiallyinconditionsofhighloadandpoor-qualitylinks(suchasinjammingscenarios).
Wemeasurethenetwork-widecon-vergencetime,i.
e.
,thetimeelapsedfromthemomentthatweactivatethejammeruntilalllegitimatedeviceshavead-justedtheirCCAthresholdasperourpowercontrolscheme.
First,weperformmeasurementsonamulti-hopmeshtopol-ogyconsistingof5APsand10clients(2clientsperAP).
Inordertohaveanideaaboutwhethertheobservedconver-gencetimeissignicant,wealsoperformexperimentswith-outjammers,whereinwemanuallyinvokethepowercontrolmodulethroughauser-levelsocketinterfaceononeoftheAPs.
Weobservethattheconvergencetimeforthespecicsettingisapproximately1.
2sec.
Then,weactivateacontin-uouslytransmittingdeceptivejammerinacloseproximityto2neighborAPs(MRCisdisabled;thejammeraectsonlythe2APs).
Table5containsvariousaverageconvergencetimesforthespecicsettingandfordierentPJvalues.
PJ(dB)Convergencetime(sec)11.
822.
432.
843.
5Table5:Averageconvergencetimes(insec)fordif-ferentPJvalues.
Weobservethatalthoughtheconvergencetimeincreasesduetojamming,itstillremainsshort.
Furthermore,weperformextensiveexperimentswith8APs,19clientsand4balancedjammerswithPJ=3dBm,alluniformlyde-ployed.
Weobservethatinitsdistributedformthepowercontrolmoduleconvergesinapproximately16secinournetwork-wideexperiments.
Althoughonemayexpectdier-ent(lowerorhigher)convergencetimeswithdierenthard-ware/softwareand/ormobilejammers,theseresultsshowthatinastatictopologythepowercontrolmoduleconvergesrelativelyquicklyinpracticalsettings.
7.
CONCLUSIONSWedesign,implementandevaluateARES,ananti-jammingsystemfor802.
11networks.
AREShasbeenbuiltbasedonobservationsfromextensivemeasurementsonanindoortestbedinthepresenceofrandomjammers,andisprimar-ilycomposedoftwomodules.
ThepowercontrolmoduletunestheCCAthresholdsinordertoallowthetransmis-sionandcaptureoflegitimatepacketsinthepresenceofthejammer'ssignals,totheextentpossible.
Theratecontrolmoduledecidesbetweenxedoradaptive-rateassignment.
WedemonstratetheeectivenessofARESinthreedierentdeployments(a)a802.
11nbasedMIMOWLAN,(b)an802.
11anetworkinfestedwithmobilejammers,and(c)a802.
11aWLANwithuplinkTCPtrac.
AREScanbeusedinconjunctionwithotherjammingmitigationtechniques(suchasfrequencyhoppingordirectionalantennas).
Over-all,theapplicationofARESleadstosignicantperformancebenetsinjammingenvironments.
Acknowledgments:WethankRalinkCorp.
forpro-vidingthesourceoftheRT2860APdriver,andDr.
Kon-stantinaPapagiannakifromIntelResearchforprovidingtheprototypeversionoftheipw2200driver.
8.
REFERENCES[1]SESPjammers.
http://www.
sesp.
com/.
[2]Jammingattackathackerconference.
http://ndarticles.
com/p/articles/mim0EIN/is2005August2/ain14841565.
[3]Techworldnews.
http://www.
techworld.
com/mobility/news/index.
cfmnewsid=10941.
[4]RFJammingAttack.
http://manageengine.
adventnet.
com/products/wi-manager/rfjamming-attack.
html.
[5]V.
Navda,A.
Bohra,S.
Ganguly,andD.
Rubenstein.
UsingChannelHoppingtoIncrease802.
11ResiliencetoJammingAttacks.
InIEEEINFOCOMmini-conference,2007.
[6]W.
Hu,T.
Wood,W.
Trappe,andY.
Zhang.
ChannelSurngandSpatialRetreats:DefensesAgainstWirelessDenialofService.
InACMWorkshoponWirelessSecurity,2004.
[7]ISMWide-bandJammers.
http://69.
6.
206.
229/e-commerce-solutions-catalog1.
0.
4.
html.
[8]ISA:Usersfearwirelessnetworksforcontrol.
http://lists.
jammed.
com/ISN/2007/05/0122.
html.
[9]K.
Pelechrinis,C.
Koufogiannakis,andS.
V.
Krishnamurthy.
GamingtheJammer:IsFrequencyHoppingEectiveInWiOpt,June2009.
[10]V.
Mhatre,K.
Papagiannaki,andF.
Baccelli.
InterferenceMitigationthroughPowerControlinHighDensity802.
11WLANs.
InIEEEINFOCOM,2007.
[11]J.
Bicket.
Bit-rateSelectioninWirelessNetworks.
InMSThesis,Dept.
ofElectr.
Engin.
andComp.
Science,MIT,2005.
[12]OnoeRateControl.
http://madwi.
org/browser/trunk/athrate/onoe.
[13]S.
Pal,S.
R.
Kundu,K.
Basu,andS.
K.
Das.
IEEE802.
11RateControlAlgorithms:ExperimentationandPerformanceEvaluationinInfrastructureMode.
InPAM,2006.
[14]W.
Xu,W.
Trappe,Y.
Zhang,andT.
Wood.
TheFeasibilityofLaunchingandDetectingJammingAttacksinWirelessNetworks.
InACMMOBIHOC,2005.
[15]R.
Gummadi,D.
Wetheral,B.
Greenstein,andS.
Seshan.
UnderstandingandMitigatingtheImpactofRFInterferenceon802.
11Networks.
InACMSIGCOMM,2007.
[16]W.
Hu,K.
Ma,W.
Trappe,andY.
Zhang.
JammingSensorNetworks:AttacksandDefenseStrategies.
InIEEENetwork,May/June2006.
[17]G.
LinandG.
Noubir.
OnLinkLayerDenialofServiceinDataWirelessLANs.
InWirelessCommunicationsandMobileComputing,May2003.
[18]G.
NoubirandG.
Lin.
Low-powerDoSAttacksinDataWirelessLANsandCountermeasures.
InACMMOBIHOC(poster),2003.
[19]G.
Noubir.
OnConnectivityinAdHocNetworkunderJammingUsingDirectionalAntennasandMobility.
InWired/WirelessInternetCommunications,Vol.
2957/2004,pp.
186-200,2004.
[20]UCRWirelesstestbed.
http://networks.
cs.
ucr.
edu/testbed.
[21]TheMadWiFidriver.
http://madwi-project.
org/.
[22]RT2860wirelessdriver.
http://www.
ralinktech.
com/ralink/Home/Support/Linux.
html.
[23]D.
S.
J.
DeCouto,D.
Aguayo,J.
Bicket,andR.
Morris.
AHighThroughputPathMetricforMultiHopWirelessRouting.
InACMMOBICOM,2003.
[24]J.
C.
ChenandJ.
M.
Gilbert.
MeasuredPerformanceof5-GHz802.
11aWirelessLANSystems.
InAtherosComm.
WhitePaper,August2001.
[25]I.
Broustis,K.
Papagiannaki,S.
V.
Krishnamurthy,M.
Faloutsos,andV.
Mhatre.
MDG:Measurement-DrivenGuidelinesfor802.
11WLANDesign.
InACMMOBICOM,2007.
[26]SampleRateBug.
http://madwi.
org/ticket/989.
[27]S.
Zvanovec,P.
Pechac,andM.
Klepal.
WirelessLANNetworksDesign:SiteSyrveyorPropagationModelsInRadioengineering,Vol.
12,No.
4,Dec.
2003.
[28]H.
Jafarkhani.
Space-TimeCoding:TheoryandPractice.
CambridgeUniversityPress,2005.
[29]S.
Vasudevanetal.
FacilitatingAccessPointSelectioninIEEE802.
11WirelessNetworks.
InACMIMC,2005.